Deep learning-based framework for real-time upper limb motion intention classification using combined bio-signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Engineering

Background:

• Advanced prosthetic limbs require intuitive control interfaces.
• Decoding human intention for upper limb movement is crucial for prosthetic functionality.
• Integrating multiple biosignals can improve the accuracy of intention decoding.

Purpose of the Study:

• To develop a real-time neuro-machine interface for controlling prosthetic arms.
• To decode human intention for upper limb motions using combined sEMG and fNIRS signals.
• To evaluate the framework's performance in trans-humeral amputees.

Main Methods:

• A novel framework integrating surface electromyogram (sEMG) and functional near-infrared spectroscopy (fNIRS) bio-signals.
• Convolutional Neural Networks (CNNs) were employed for signal training and intention decoding.
• fNIRS signals from the motor cortex and sEMG from bicep muscles were recorded simultaneously for eight upper limb movements.
• Feature extraction included peak, minimum, and mean ΔHbO and ΔHbR values (fNIRS) and wavelength, peak, and mean (sEMG) within specific moving windows.

Main Results:

• The framework successfully decoded eight distinct upper limb movements.
• An enhanced average accuracy of 94.5% was achieved in classifying intended motions.
• The selected features from both sEMG and fNIRS proved effective for intention decoding.

Conclusions:

• The proposed framework demonstrates a high potential for real-time control of prosthetic arms.
• The integration of sEMG and fNIRS offers a robust approach for neuro-machine interfaces.
• This research validates a promising methodology for improving prosthetic arm functionality and user control.